From 9ba488620d188c8a732c7e38dd19279b512ee9ad Mon Sep 17 00:00:00 2001
From: NamH <NamNh0122@gmail.com>
Date: Fri, 27 Dec 2024 08:27:44 +0700
Subject: [PATCH 1/3] fix: not create new folder if is registering paths
 (#1828)

---
 engine/services/engine_service.cc  | 7 ++++---
 engine/utils/file_manager_utils.cc | 5 +++--
 engine/utils/file_manager_utils.h  | 3 ++-
 3 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/engine/services/engine_service.cc b/engine/services/engine_service.cc
index c8f4c180c..79b50df36 100644
--- a/engine/services/engine_service.cc
+++ b/engine/services/engine_service.cc
@@ -132,8 +132,8 @@ cpp::result<bool, std::string> EngineService::UnzipEngine(
           CTL_INF("Found cuda variant, extract it");
           found_cuda = true;
           // extract binary
-          auto cuda_path =
-              file_manager_utils::GetCudaToolkitPath(NormalizeEngine(engine));
+          auto cuda_path = file_manager_utils::GetCudaToolkitPath(
+              NormalizeEngine(engine), true);
           archive_utils::ExtractArchive(path + "/" + cf, cuda_path.string(),
                                         true);
         }
@@ -434,7 +434,8 @@ cpp::result<bool, std::string> EngineService::DownloadCuda(
   }};
 
   auto on_finished = [engine](const DownloadTask& finishedTask) {
-    auto engine_path = file_manager_utils::GetCudaToolkitPath(engine);
+    auto engine_path = file_manager_utils::GetCudaToolkitPath(engine, true);
+
     archive_utils::ExtractArchive(finishedTask.items[0].localPath.string(),
                                   engine_path.string());
     try {
diff --git a/engine/utils/file_manager_utils.cc b/engine/utils/file_manager_utils.cc
index 338abadac..aee65020c 100644
--- a/engine/utils/file_manager_utils.cc
+++ b/engine/utils/file_manager_utils.cc
@@ -289,13 +289,14 @@ std::filesystem::path GetModelsContainerPath() {
   return models_container_path;
 }
 
-std::filesystem::path GetCudaToolkitPath(const std::string& engine) {
+std::filesystem::path GetCudaToolkitPath(const std::string& engine,
+                                         bool create_if_not_exist) {
   auto engine_path = getenv("ENGINE_PATH")
                          ? std::filesystem::path(getenv("ENGINE_PATH"))
                          : GetCortexDataPath();
 
   auto cuda_path = engine_path / "engines" / engine / "deps";
-  if (!std::filesystem::exists(cuda_path)) {
+  if (create_if_not_exist && !std::filesystem::exists(cuda_path)) {
     std::filesystem::create_directories(cuda_path);
   }
 
diff --git a/engine/utils/file_manager_utils.h b/engine/utils/file_manager_utils.h
index 91102d002..059fe6ae3 100644
--- a/engine/utils/file_manager_utils.h
+++ b/engine/utils/file_manager_utils.h
@@ -45,7 +45,8 @@ void CreateDirectoryRecursively(const std::string& path);
 
 std::filesystem::path GetModelsContainerPath();
 
-std::filesystem::path GetCudaToolkitPath(const std::string& engine);
+std::filesystem::path GetCudaToolkitPath(const std::string& engine,
+                                         bool create_if_not_exist = false);
 
 std::filesystem::path GetEnginesContainerPath();
 

From 3cc1e71520c6482a2076d1a601d2f489e53d788d Mon Sep 17 00:00:00 2001
From: vansangpfiev <vansangpfiev@gmail.com>
Date: Thu, 26 Dec 2024 09:25:29 +0700
Subject: [PATCH 2/3] fix: forward start model parameters (#1825)

Co-authored-by: vansangpfiev <sang@jan.ai>
---
 engine/controllers/models.cc               | 53 ++++++--------------
 engine/services/model_service.cc           | 35 +++++++------
 engine/services/model_service.h            | 19 +------
 engine/test/components/test_json_helper.cc | 58 ++++++++++++++++++++++
 engine/utils/json_helper.h                 | 24 +++++++++
 5 files changed, 116 insertions(+), 73 deletions(-)

diff --git a/engine/controllers/models.cc b/engine/controllers/models.cc
index 59793b2a6..1c33ab1dc 100644
--- a/engine/controllers/models.cc
+++ b/engine/controllers/models.cc
@@ -488,55 +488,31 @@ void Models::StartModel(
   if (!http_util::HasFieldInReq(req, callback, "model"))
     return;
   auto model_handle = (*(req->getJsonObject())).get("model", "").asString();
-  StartParameterOverride params_override;
-  if (auto& o = (*(req->getJsonObject()))["prompt_template"]; !o.isNull()) {
-    params_override.custom_prompt_template = o.asString();
-  }
-
-  if (auto& o = (*(req->getJsonObject()))["cache_enabled"]; !o.isNull()) {
-    params_override.cache_enabled = o.asBool();
-  }
-
-  if (auto& o = (*(req->getJsonObject()))["ngl"]; !o.isNull()) {
-    params_override.ngl = o.asInt();
-  }
-
-  if (auto& o = (*(req->getJsonObject()))["n_parallel"]; !o.isNull()) {
-    params_override.n_parallel = o.asInt();
-  }
-
-  if (auto& o = (*(req->getJsonObject()))["ctx_len"]; !o.isNull()) {
-    params_override.ctx_len = o.asInt();
-  }
-
-  if (auto& o = (*(req->getJsonObject()))["cache_type"]; !o.isNull()) {
-    params_override.cache_type = o.asString();
-  }
 
+  std::optional<std::string> mmproj;
   if (auto& o = (*(req->getJsonObject()))["mmproj"]; !o.isNull()) {
-    params_override.mmproj = o.asString();
+    mmproj = o.asString();
   }
 
+  auto bypass_llama_model_path = false;
   // Support both llama_model_path and model_path for backward compatible
   // model_path has higher priority
   if (auto& o = (*(req->getJsonObject()))["llama_model_path"]; !o.isNull()) {
-    params_override.model_path = o.asString();
+    auto model_path = o.asString();
     if (auto& mp = (*(req->getJsonObject()))["model_path"]; mp.isNull()) {
       // Bypass if model does not exist in DB and llama_model_path exists
-      if (std::filesystem::exists(params_override.model_path.value()) &&
+      if (std::filesystem::exists(model_path) &&
           !model_service_->HasModel(model_handle)) {
         CTL_INF("llama_model_path exists, bypass check model id");
-        params_override.bypass_llama_model_path = true;
+        bypass_llama_model_path = true;
       }
     }
   }
 
-  if (auto& o = (*(req->getJsonObject()))["model_path"]; !o.isNull()) {
-    params_override.model_path = o.asString();
-  }
+  auto bypass_model_check = (mmproj.has_value() || bypass_llama_model_path);
 
   auto model_entry = model_service_->GetDownloadedModel(model_handle);
-  if (!model_entry.has_value() && !params_override.bypass_model_check()) {
+  if (!model_entry.has_value() && !bypass_model_check) {
     Json::Value ret;
     ret["message"] = "Cannot find model: " + model_handle;
     auto resp = cortex_utils::CreateCortexHttpJsonResponse(ret);
@@ -544,9 +520,8 @@ void Models::StartModel(
     callback(resp);
     return;
   }
-  std::string engine_name = params_override.bypass_model_check()
-                                ? kLlamaEngine
-                                : model_entry.value().engine;
+  std::string engine_name =
+      bypass_model_check ? kLlamaEngine : model_entry.value().engine;
   auto engine_validate = engine_service_->IsEngineReady(engine_name);
   if (engine_validate.has_error()) {
     Json::Value ret;
@@ -565,7 +540,9 @@ void Models::StartModel(
     return;
   }
 
-  auto result = model_service_->StartModel(model_handle, params_override);
+  auto result = model_service_->StartModel(
+      model_handle, *(req->getJsonObject()) /*params_override*/,
+      bypass_model_check);
   if (result.has_error()) {
     Json::Value ret;
     ret["message"] = result.error();
@@ -668,7 +645,7 @@ void Models::AddRemoteModel(
 
   auto model_handle = (*(req->getJsonObject())).get("model", "").asString();
   auto engine_name = (*(req->getJsonObject())).get("engine", "").asString();
- 
+
   auto engine_validate = engine_service_->IsEngineReady(engine_name);
   if (engine_validate.has_error()) {
     Json::Value ret;
@@ -687,7 +664,7 @@ void Models::AddRemoteModel(
     callback(resp);
     return;
   }
-  
+
   config::RemoteModelConfig model_config;
   model_config.LoadFromJson(*(req->getJsonObject()));
   cortex::db::Models modellist_utils_obj;
diff --git a/engine/services/model_service.cc b/engine/services/model_service.cc
index ce83152c4..b79894ca1 100644
--- a/engine/services/model_service.cc
+++ b/engine/services/model_service.cc
@@ -748,19 +748,28 @@ cpp::result<void, std::string> ModelService::DeleteModel(
 }
 
 cpp::result<StartModelResult, std::string> ModelService::StartModel(
-    const std::string& model_handle,
-    const StartParameterOverride& params_override) {
+    const std::string& model_handle, const Json::Value& params_override,
+    bool bypass_model_check) {
   namespace fs = std::filesystem;
   namespace fmu = file_manager_utils;
   cortex::db::Models modellist_handler;
   config::YamlHandler yaml_handler;
+  std::optional<std::string> custom_prompt_template;
+  std::optional<int> ctx_len;
+  if (auto& o = params_override["prompt_template"]; !o.isNull()) {
+    custom_prompt_template = o.asString();
+  }
+
+  if (auto& o = params_override["ctx_len"]; !o.isNull()) {
+    ctx_len = o.asInt();
+  }
 
   try {
     constexpr const int kDefautlContextLength = 8192;
     int max_model_context_length = kDefautlContextLength;
     Json::Value json_data;
     // Currently we don't support download vision models, so we need to bypass check
-    if (!params_override.bypass_model_check()) {
+    if (!bypass_model_check) {
       auto model_entry = modellist_handler.GetModelInfo(model_handle);
       if (model_entry.has_error()) {
         CTL_WRN("Error: " + model_entry.error());
@@ -838,29 +847,19 @@ cpp::result<StartModelResult, std::string> ModelService::StartModel(
     }
 
     json_data["model"] = model_handle;
-    if (auto& cpt = params_override.custom_prompt_template;
-        !cpt.value_or("").empty()) {
+    if (auto& cpt = custom_prompt_template; !cpt.value_or("").empty()) {
       auto parse_prompt_result = string_utils::ParsePrompt(cpt.value());
       json_data["system_prompt"] = parse_prompt_result.system_prompt;
       json_data["user_prompt"] = parse_prompt_result.user_prompt;
       json_data["ai_prompt"] = parse_prompt_result.ai_prompt;
     }
 
-#define ASSIGN_IF_PRESENT(json_obj, param_override, param_name) \
-  if (param_override.param_name) {                              \
-    json_obj[#param_name] = param_override.param_name.value();  \
-  }
+    json_helper::MergeJson(json_data, params_override);
 
-    ASSIGN_IF_PRESENT(json_data, params_override, cache_enabled);
-    ASSIGN_IF_PRESENT(json_data, params_override, ngl);
-    ASSIGN_IF_PRESENT(json_data, params_override, n_parallel);
-    ASSIGN_IF_PRESENT(json_data, params_override, cache_type);
-    ASSIGN_IF_PRESENT(json_data, params_override, mmproj);
-    ASSIGN_IF_PRESENT(json_data, params_override, model_path);
-#undef ASSIGN_IF_PRESENT
-    if (params_override.ctx_len) {
+    // Set the latest ctx_len
+    if (ctx_len) {
       json_data["ctx_len"] =
-          std::min(params_override.ctx_len.value(), max_model_context_length);
+          std::min(ctx_len.value(), max_model_context_length);
     }
     CTL_INF(json_data.toStyledString());
     auto may_fallback_res = MayFallbackToCpu(json_data["model_path"].asString(),
diff --git a/engine/services/model_service.h b/engine/services/model_service.h
index e2638fd1f..35c5ba539 100644
--- a/engine/services/model_service.h
+++ b/engine/services/model_service.h
@@ -18,21 +18,6 @@ struct ModelPullInfo {
   std::string download_url;
 };
 
-struct StartParameterOverride {
-  std::optional<bool> cache_enabled;
-  std::optional<int> ngl;
-  std::optional<int> n_parallel;
-  std::optional<int> ctx_len;
-  std::optional<std::string> custom_prompt_template;
-  std::optional<std::string> cache_type;
-  std::optional<std::string> mmproj;
-  std::optional<std::string> model_path;
-  bool bypass_llama_model_path = false;
-  bool bypass_model_check() const {
-    return mmproj.has_value() || bypass_llama_model_path;
-  }
-};
-
 struct StartModelResult {
   bool success;
   std::optional<std::string> warning;
@@ -78,8 +63,8 @@ class ModelService {
   cpp::result<void, std::string> DeleteModel(const std::string& model_handle);
 
   cpp::result<StartModelResult, std::string> StartModel(
-      const std::string& model_handle,
-      const StartParameterOverride& params_override);
+      const std::string& model_handle, const Json::Value& params_override,
+      bool bypass_model_check);
 
   cpp::result<bool, std::string> StopModel(const std::string& model_handle);
 
diff --git a/engine/test/components/test_json_helper.cc b/engine/test/components/test_json_helper.cc
index cb3f4683a..ba5e27165 100644
--- a/engine/test/components/test_json_helper.cc
+++ b/engine/test/components/test_json_helper.cc
@@ -33,3 +33,61 @@ TEST(ParseJsonStringTest, EmptyString) {
     
     EXPECT_TRUE(result.isNull()); 
 }
+
+TEST(MergeJsonTest, MergeSimpleObjects) {
+    Json::Value json1, json2;
+    json1["name"] = "John";
+    json1["age"] = 30;
+
+    json2["age"] = 31;
+    json2["email"] = "john@example.com";
+
+    json_helper::MergeJson(json1, json2);
+
+    Json::Value expected;
+    expected["name"] = "John";
+    expected["age"] = 31;
+    expected["email"] = "john@example.com";
+
+    EXPECT_EQ(json1, expected);
+}
+
+TEST(MergeJsonTest, MergeNestedObjects) {
+    Json::Value json1, json2;
+    json1["person"]["name"] = "John";
+    json1["person"]["age"] = 30;
+
+    json2["person"]["age"] = 31;
+    json2["person"]["email"] = "john@example.com";
+
+    json_helper::MergeJson(json1, json2);
+
+    Json::Value expected;
+    expected["person"]["name"] = "John";
+    expected["person"]["age"] = 31;
+    expected["person"]["email"] = "john@example.com";
+
+    EXPECT_EQ(json1, expected);
+}
+
+TEST(MergeJsonTest, MergeArrays) {
+    Json::Value json1, json2;
+    json1["hobbies"] = Json::Value(Json::arrayValue);
+    json1["hobbies"].append("reading");
+    json1["hobbies"].append("painting");
+
+    json2["hobbies"] = Json::Value(Json::arrayValue);
+    json2["hobbies"].append("hiking");
+    json2["hobbies"].append("painting");
+
+    json_helper::MergeJson(json1, json2);
+
+    Json::Value expected;
+    expected["hobbies"] = Json::Value(Json::arrayValue);
+    expected["hobbies"].append("reading");
+    expected["hobbies"].append("painting");
+    expected["hobbies"].append("hiking");
+    expected["hobbies"].append("painting");
+
+    EXPECT_EQ(json1, expected);
+}
diff --git a/engine/utils/json_helper.h b/engine/utils/json_helper.h
index 82f994751..3b08651c4 100644
--- a/engine/utils/json_helper.h
+++ b/engine/utils/json_helper.h
@@ -16,4 +16,28 @@ inline std::string DumpJsonString(const Json::Value& json) {
   builder["indentation"] = "";
   return Json::writeString(builder, json);
 }
+
+inline void MergeJson(Json::Value& target, const Json::Value& source) {
+  for (const auto& member : source.getMemberNames()) {
+    if (target.isMember(member)) {
+      // If the member exists in both objects, recursively merge the values
+      if (target[member].type() == Json::objectValue &&
+          source[member].type() == Json::objectValue) {
+        MergeJson(target[member], source[member]);
+      } else if (target[member].type() == Json::arrayValue &&
+                 source[member].type() == Json::arrayValue) {
+        // If the member is an array in both objects, merge the arrays
+        for (const auto& value : source[member]) {
+          target[member].append(value);
+        }
+      } else {
+        // Otherwise, overwrite the value in the target with the value from the source
+        target[member] = source[member];
+      }
+    } else {
+      // If the member doesn't exist in the target, add it
+      target[member] = source[member];
+    }
+  }
+}
 }  // namespace json_helper

From 5825412b362198c3dd7083b7dc33717cc596dbba Mon Sep 17 00:00:00 2001
From: vansangpfiev <vansangpfiev@gmail.com>
Date: Wed, 8 Jan 2025 09:19:08 +0700
Subject: [PATCH 3/3] fix: build linux arm (#1806)

* fix: build linux arm

* feat: cicd arm64

* fix: build linux arm

* feat: cicd arm64

* fix: e2e test linux arm64

* fix: select linux arm64

* chore: e2e tests

* fix: ci linux arm

* fix: ci correct artifact name

* fix: linux installer arm64

---------

Co-authored-by: Hien To <tominhhien97@gmail.com>
Co-authored-by: sangjanai <sang@jan.ai>
Co-authored-by: Service Account <service@jan.ai>
---
 .github/workflows/beta-build.yml              | 26 +++++-
 .github/workflows/cortex-cpp-quality-gate.yml | 36 +++++++++
 .github/workflows/nightly-build.yml           | 21 ++++-
 .github/workflows/stable-build.yml            | 18 ++++-
 ...linux-x64.yml => template-build-linux.yml} | 80 ++++++++++++-------
 engine/Makefile                               |  3 +-
 engine/e2e-test/test_api_engine.py            |  4 +-
 engine/e2e-test/test_api_engine_update.py     |  2 +-
 engine/e2e-test/test_api_model.py             |  1 +
 engine/e2e-test/test_cli_engine_install.py    |  4 +-
 .../test_cli_engine_install_nightly.py        |  1 +
 engine/e2e-test/test_cli_engine_uninstall.py  |  2 +-
 engine/services/download_service.cc           | 10 +--
 engine/templates/linux/control                |  2 +-
 engine/templates/linux/create_deb.sh          |  4 +-
 engine/templates/linux/create_deb_local.sh    |  3 +-
 engine/templates/linux/install.sh             | 29 ++++---
 .../components/test_engine_matcher_utils.cc   | 13 +++
 engine/utils/curl_utils.cc                    | 77 +++++++++++-------
 engine/utils/curl_utils.h                     |  7 +-
 engine/utils/engine_matcher_utils.h           |  5 ++
 21 files changed, 255 insertions(+), 93 deletions(-)
 rename .github/workflows/{template-build-linux-x64.yml => template-build-linux.yml} (66%)

diff --git a/.github/workflows/beta-build.yml b/.github/workflows/beta-build.yml
index c5c09dcb5..bdc277231 100644
--- a/.github/workflows/beta-build.yml
+++ b/.github/workflows/beta-build.yml
@@ -67,7 +67,7 @@ jobs:
       cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
           
   build-linux-x64:
-    uses: ./.github/workflows/template-build-linux-x64.yml
+    uses: ./.github/workflows/template-build-linux.yml
     secrets: inherit
     needs: [get-update-version, create-draft-release, get-cortex-llamacpp-latest-version]
     with:
@@ -79,6 +79,22 @@ jobs:
       channel: beta
       upload_url: ${{ needs.create-draft-release.outputs.upload_url }}
       cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
+      arch: amd64
+
+  build-linux-arm64:
+    uses: ./.github/workflows/template-build-linux.yml
+    secrets: inherit
+    needs: [get-update-version, create-draft-release, get-cortex-llamacpp-latest-version]
+    with:
+      ref: ${{ github.ref }}
+      public_provider: github
+      new_version: ${{ needs.get-update-version.outputs.new_version }}
+      runs-on: ubuntu-2004-arm64
+      cmake-flags: "-DCORTEX_VARIANT=beta -DCORTEX_CPP_VERSION='v${{ needs.get-update-version.outputs.new_version }}' -DCMAKE_TOOLCHAIN_FILE=/home/runner/actions-runner/_work/cortex.cpp/cortex.cpp/engine/vcpkg/scripts/buildsystems/vcpkg.cmake"
+      channel: beta
+      upload_url: ${{ needs.create-draft-release.outputs.upload_url }}
+      cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
+      arch: arm64
 
   build-docker-x64:
     uses: ./.github/workflows/template-build-docker-x64.yml
@@ -111,7 +127,7 @@ jobs:
             GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
   noti-discord:
-    needs: [get-update-version, create-draft-release, build-macos, build-windows-x64, build-linux-x64, update_release]
+    needs: [get-update-version, create-draft-release, build-macos, build-windows-x64, build-linux-x64, build-linux-arm64, update_release]
     runs-on: ubuntu-latest
     permissions:
       contents: write
@@ -136,9 +152,13 @@ jobs:
               - Network Installer: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-mac-universal-network-installer.pkg
               - Local Installer: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-mac-universal-local-installer.pkg
               - Binary: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-mac-universal.tar.gz
-            - Linux Deb:
+            - Linux amd64 Deb:
               - Network Installer: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-linux-amd64-network-installer.deb
               - Local Installer: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-linux-amd64-local-installer.deb
               - Binary: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-linux-amd64.tar.gz
+            - Linux amd64 Deb:
+              - Network Installer: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-linux-arm64-network-installer.deb
+              - Local Installer: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-linux-arm64-local-installer.deb
+              - Binary: https://github.com/janhq/cortex.cpp/releases/download/v${{ env.VERSION }}/cortex-${{ env.VERSION }}-linux-arm64.tar.gz
             - Docker: menloltd/cortex:beta-${{ env.VERSION }}
             - Github Release: https://github.com/janhq/cortex.cpp/releases/tag/v${{ env.VERSION }}
\ No newline at end of file
diff --git a/.github/workflows/cortex-cpp-quality-gate.yml b/.github/workflows/cortex-cpp-quality-gate.yml
index 8a76e4669..fd98930a1 100644
--- a/.github/workflows/cortex-cpp-quality-gate.yml
+++ b/.github/workflows/cortex-cpp-quality-gate.yml
@@ -20,6 +20,12 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          - os: "linux"
+            name: "arm64"
+            runs-on: "ubuntu-2004-arm64"
+            cmake-flags: "-DCORTEX_CPP_VERSION=${{github.event.pull_request.head.sha}} -DCMAKE_BUILD_TEST=ON -DCMAKE_TOOLCHAIN_FILE=vcpkg/scripts/buildsystems/vcpkg.cmake"
+            build-deps-cmake-flags: ""
+            ccache-dir: ""
           - os: "linux"
             name: "amd64"
             runs-on: "ubuntu-20-04-cuda-12-0"
@@ -52,6 +58,7 @@ jobs:
           submodules: recursive
 
       - name: use python
+        continue-on-error: true
         uses: actions/setup-python@v5
         with:
           python-version: "3.10"
@@ -90,15 +97,44 @@ jobs:
           AWS_DEFAULT_REGION: "${{ secrets.MINIO_REGION }}"
 
       - name: Configure vcpkg
+        if: runner.os != 'Linux'
+        run: |
+          cd engine
+          make configure-vcpkg
+
+      - name: Configure vcpkg linux amd64
+        if: runner.os != 'Linux'
+        run: |
+          cd engine
+          make configure-vcpkg
+
+      - name: Configure vcpkg linux arm64
+        if: runner.os == 'Linux'
         run: |
           cd engine
+          # Set env if arch is arm64
+          if [ "${{ matrix.name }}" == "arm64" ]; then
+            sudo apt install ninja-build pkg-config -y
+            export VCPKG_FORCE_SYSTEM_BINARIES=1
+          fi
           make configure-vcpkg
 
       - name: Build
+        if: runner.os != 'Linux'
         run: |
           cd engine
           make build CMAKE_EXTRA_FLAGS="${{ matrix.cmake-flags }}" BUILD_DEPS_CMAKE_EXTRA_FLAGS="${{ matrix.build-deps-cmake-flags }}"
 
+      - name: Build
+        if: runner.os == 'Linux'
+        run: |
+          cd engine
+          if [ "${{ matrix.name }}" == "arm64" ]; then
+            export VCPKG_FORCE_SYSTEM_BINARIES=1
+          fi
+          make build CMAKE_EXTRA_FLAGS="${{ matrix.cmake-flags }}" BUILD_DEPS_CMAKE_EXTRA_FLAGS="${{ matrix.build-deps-cmake-flags }}"
+
+
       - name: Run setup config
         run: |
           cd engine
diff --git a/.github/workflows/nightly-build.yml b/.github/workflows/nightly-build.yml
index 9a31ef5ff..1f076dc97 100644
--- a/.github/workflows/nightly-build.yml
+++ b/.github/workflows/nightly-build.yml
@@ -74,7 +74,7 @@ jobs:
       cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
           
   build-linux-x64:
-    uses: ./.github/workflows/template-build-linux-x64.yml
+    uses: ./.github/workflows/template-build-linux.yml
     secrets: inherit
     needs: [get-update-version, set-public-provider, get-cortex-llamacpp-latest-version]
     with:
@@ -85,11 +85,26 @@ jobs:
       cmake-flags: "-DCORTEX_VARIANT=nightly -DCORTEX_CPP_VERSION='v${{ needs.get-update-version.outputs.new_version }}' -DCMAKE_TOOLCHAIN_FILE=/home/runner/actions-runner/_work/cortex.cpp/cortex.cpp/engine/vcpkg/scripts/buildsystems/vcpkg.cmake"
       channel: nightly
       cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
+      arch: amd64
+
+  build-linux-arm64:
+    uses: ./.github/workflows/template-build-linux.yml
+    secrets: inherit
+    needs: [get-update-version, set-public-provider, get-cortex-llamacpp-latest-version]
+    with:
+      ref: ${{ needs.set-public-provider.outputs.ref }}
+      public_provider: ${{ needs.set-public-provider.outputs.public_provider }}
+      new_version: ${{ needs.get-update-version.outputs.new_version }}
+      runs-on: ubuntu-2004-arm64
+      cmake-flags: "-DCORTEX_VARIANT=nightly -DCORTEX_CPP_VERSION='v${{ needs.get-update-version.outputs.new_version }}' -DCMAKE_TOOLCHAIN_FILE=/home/runner/actions-runner/_work/cortex.cpp/cortex.cpp/engine/vcpkg/scripts/buildsystems/vcpkg.cmake"
+      channel: nightly
+      cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
+      arch: arm64
 
   update-latest-version:
     runs-on: ubuntu-latest
     if: needs.set-public-provider.outputs.public_provider == 'aws-s3'
-    needs: [get-update-version, set-public-provider, build-linux-x64, build-macos, build-windows-x64, get-cortex-llamacpp-latest-version]
+    needs: [get-update-version, set-public-provider, build-linux-x64, build-linux-arm64, build-macos, build-windows-x64, get-cortex-llamacpp-latest-version]
     steps:
       - name: Update latest version
         id: update-latest-version
@@ -100,9 +115,11 @@ jobs:
           aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/mac-universal-cortex-nightly.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/mac-arm64/cortex-nightly.tar.gz
           aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/mac-universal-cortex-nightly.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/mac-universal/cortex-nightly.tar.gz
           aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/linux-amd64-cortex-nightly.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/linux-amd64/cortex-nightly.tar.gz
+          aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/linux-arm64-cortex-nightly.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/linux-arm64/cortex-nightly.tar.gz
           aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/windows-amd64-cortex-nightly.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/windows-amd64/cortex-nightly.tar.gz
           aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/cortex-mac-universal-network-installer.pkg s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/mac-universal/cortex-mac-universal-network-installer.pkg
           aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/cortex-linux-amd64-network-installer.deb s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/linux-amd64/cortex-linux-amd64-network-installer.deb
+          aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/cortex-linux-arm64-network-installer.deb s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/linux-arm64/cortex-linux-arm64-network-installer.deb
           aws s3 cp s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/cortex-windows-amd64-network-installer.exe s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/latest/windows-amd64/cortex-windows-amd64-network-installer.exe
 
         env:
diff --git a/.github/workflows/stable-build.yml b/.github/workflows/stable-build.yml
index 2b0523771..b05df983d 100644
--- a/.github/workflows/stable-build.yml
+++ b/.github/workflows/stable-build.yml
@@ -67,7 +67,7 @@ jobs:
       cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
           
   build-linux-x64:
-    uses: ./.github/workflows/template-build-linux-x64.yml
+    uses: ./.github/workflows/template-build-linux.yml
     secrets: inherit
     needs: [get-update-version, create-draft-release, get-cortex-llamacpp-latest-version]
     with:
@@ -79,6 +79,22 @@ jobs:
       channel: stable
       upload_url: ${{ needs.create-draft-release.outputs.upload_url }}
       cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
+      arch: amd64
+
+  build-linux-arm64:
+    uses: ./.github/workflows/template-build-linux.yml
+    secrets: inherit
+    needs: [get-update-version, create-draft-release, get-cortex-llamacpp-latest-version]
+    with:
+      ref: ${{ github.ref }}
+      public_provider: github
+      new_version: ${{ needs.get-update-version.outputs.new_version }}
+      runs-on: ubuntu-2004-arm64
+      cmake-flags: "-DCORTEX_VARIANT=prod -DCORTEX_CPP_VERSION='v${{ needs.get-update-version.outputs.new_version }}' -DCMAKE_TOOLCHAIN_FILE=/home/runner/actions-runner/_work/cortex.cpp/cortex.cpp/engine/vcpkg/scripts/buildsystems/vcpkg.cmake"
+      channel: stable
+      upload_url: ${{ needs.create-draft-release.outputs.upload_url }}
+      cortex-llamacpp-version: ${{ needs.get-cortex-llamacpp-latest-version.outputs.cortex_llamacpp_latest_version }}
+      arch: arm64
 
   build-docker-x64:
     uses: ./.github/workflows/template-build-docker-x64.yml
diff --git a/.github/workflows/template-build-linux-x64.yml b/.github/workflows/template-build-linux.yml
similarity index 66%
rename from .github/workflows/template-build-linux-x64.yml
rename to .github/workflows/template-build-linux.yml
index d1ca73844..02cc3a187 100644
--- a/.github/workflows/template-build-linux-x64.yml
+++ b/.github/workflows/template-build-linux.yml
@@ -1,4 +1,4 @@
-name: build-linux-x64
+name: build-linux
 on:
   workflow_call:
     inputs:
@@ -49,6 +49,11 @@ on:
         type: string
         default: '0.0.0'
         description: 'The version of cortex-llamacpp to use for this job'
+      arch:
+        required: false
+        type: string
+        default: 'amd64'
+        description: 'The architecture to use for this job'
     secrets:
       DELTA_AWS_S3_BUCKET_NAME:
         required: false
@@ -60,7 +65,7 @@ on:
         required: false
 
 jobs:
-  build-linux-x64:
+  build-linux:
     runs-on: ${{ inputs.runs-on }}
     permissions:
       contents: write
@@ -72,6 +77,7 @@ jobs:
           submodules: 'recursive'
 
       - name: use python 3.9
+        continue-on-error: true
         uses: actions/setup-python@v4
         with:
           python-version: '3.9'
@@ -124,14 +130,24 @@ jobs:
       - name: Configure vcpkg
         run: |
           cd engine
+          # Set env if arch is arm64
+          if [ "${{ inputs.arch }}" == "arm64" ]; then
+            sudo apt install ninja-build pkg-config -y
+            export VCPKG_FORCE_SYSTEM_BINARIES=1
+          fi
           make configure-vcpkg
 
       - name: Build
         run: |
           cd engine
+          # Set env if arch is arm64
+          if [ "${{ inputs.arch }}" == "arm64" ]; then
+            export VCPKG_FORCE_SYSTEM_BINARIES=1
+          fi
           make build CMAKE_EXTRA_FLAGS="${{ inputs.cmake-flags }}" BUILD_DEPS_CMAKE_EXTRA_FLAGS="${{ inputs.build-deps-cmake-flags }}"
           
       - name: Install Python
+        continue-on-error: true
         uses: actions/setup-python@v4
         with:
           python-version: '3.10' 
@@ -145,28 +161,32 @@ jobs:
         shell: bash
         run: |
           cd engine
-          make build-installer PACKAGE_NAME="${{ steps.set-output-params.outputs.package_name }}" SOURCE_BINARY_PATH="../../cortex/${{ steps.set-output-params.outputs.destination_binary_name }}" SOURCE_BINARY_SERVER_PATH="../../cortex/${{ steps.set-output-params.outputs.destination_binary_server_name }}" VERSION=${{ inputs.new_version }} DESTINATION_BINARY_NAME="${{ steps.set-output-params.outputs.destination_binary_name }}" DESTINATION_BINARY_SERVER_NAME="${{ steps.set-output-params.outputs.destination_binary_server_name }}" DATA_FOLDER_NAME="${{ steps.set-output-params.outputs.data_folder_name }}" CONFIGURATION_FILE_NAME="${{ steps.set-output-params.outputs.configuration_file_name }}" UNINSTALLER_FILE_NAME="${{ steps.set-output-params.outputs.uninstaller_file_name }}"
+          make build-installer PACKAGE_NAME="${{ steps.set-output-params.outputs.package_name }}" SOURCE_BINARY_PATH="../../cortex/${{ steps.set-output-params.outputs.destination_binary_name }}" SOURCE_BINARY_SERVER_PATH="../../cortex/${{ steps.set-output-params.outputs.destination_binary_server_name }}" VERSION=${{ inputs.new_version }} DESTINATION_BINARY_NAME="${{ steps.set-output-params.outputs.destination_binary_name }}" DESTINATION_BINARY_SERVER_NAME="${{ steps.set-output-params.outputs.destination_binary_server_name }}" DATA_FOLDER_NAME="${{ steps.set-output-params.outputs.data_folder_name }}" CONFIGURATION_FILE_NAME="${{ steps.set-output-params.outputs.configuration_file_name }}" UNINSTALLER_FILE_NAME="${{ steps.set-output-params.outputs.uninstaller_file_name }}" ARCH="${{ inputs.arch }}"
           mv ${{ steps.set-output-params.outputs.package_name }}.deb ${{ steps.set-output-params.outputs.package_name }}-network.deb
 
       - name: Build local Installers
         run: |
           mkdir -p engine/templates/linux/dependencies
           cd engine/templates/linux/dependencies
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx-cuda-11-7.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx-cuda-12-0.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx2-cuda-11-7.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx2-cuda-12-0.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx2.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx512-cuda-11-7.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx512-cuda-12-0.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx512.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-noavx-cuda-11-7.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-noavx-cuda-12-0.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-noavx.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-vulkan.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cuda-11-7-linux-amd64.tar.gz
-          wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cuda-12-0-linux-amd64.tar.gz
+          if [ "${{ inputs.arch }}" == "amd64" ]; then
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx-cuda-11-7.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx-cuda-12-0.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx2-cuda-11-7.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx2-cuda-12-0.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx2.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx512-cuda-11-7.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx512-cuda-12-0.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-avx512.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-noavx-cuda-11-7.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-noavx-cuda-12-0.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-noavx.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-amd64-vulkan.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cuda-11-7-linux-amd64.tar.gz
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cuda-12-0-linux-amd64.tar.gz
+          else
+            wget https://github.com/janhq/cortex.llamacpp/releases/download/v${{ inputs.cortex-llamacpp-version }}/cortex.llamacpp-${{ inputs.cortex-llamacpp-version }}-linux-arm64.tar.gz
+          fi
           cd ..
 
           # Remove network package
@@ -174,7 +194,7 @@ jobs:
           rm -rf ${{ steps.set-output-params.outputs.package_name }}
           rm ${{ steps.set-output-params.outputs.package_name }}.deb
           chmod +x create_deb_local.sh
-          ./create_deb_local.sh ${{ steps.set-output-params.outputs.package_name }} ${{ inputs.new_version }} ../../cortex/${{ steps.set-output-params.outputs.destination_binary_name }} ../../cortex/${{ steps.set-output-params.outputs.destination_binary_server_name }} ${{ steps.set-output-params.outputs.destination_binary_name }} ${{ steps.set-output-params.outputs.destination_binary_server_name }} ${{ steps.set-output-params.outputs.data_folder_name }} ${{ steps.set-output-params.outputs.configuration_file_name }};
+          ./create_deb_local.sh ${{ steps.set-output-params.outputs.package_name }} ${{ inputs.new_version }} ../../cortex/${{ steps.set-output-params.outputs.destination_binary_name }} ../../cortex/${{ steps.set-output-params.outputs.destination_binary_server_name }} ${{ steps.set-output-params.outputs.destination_binary_name }} ${{ steps.set-output-params.outputs.destination_binary_server_name }} ${{ steps.set-output-params.outputs.data_folder_name }} ${{ steps.set-output-params.outputs.configuration_file_name }} ${{ inputs.arch }};
           cp ${{ steps.set-output-params.outputs.package_name }}.deb ../../${{ steps.set-output-params.outputs.package_name }}-local.deb
 
       - name: Package
@@ -185,30 +205,30 @@ jobs:
       - name: Upload Artifact
         uses: actions/upload-artifact@v4
         with:
-          name: cortex-${{ inputs.new_version }}-linux-amd64
+          name: cortex-${{ inputs.new_version }}-linux-${{ inputs.arch }}
           path: ./engine/cortex
 
       - name: Upload Artifact
         uses: actions/upload-artifact@v4
         with:
-          name: cortex-${{ inputs.new_version }}-linux-amd64-network-installer
+          name: cortex-${{ inputs.new_version }}-linux-${{ inputs.arch }}-network-installer
           path: ./engine/${{ steps.set-output-params.outputs.package_name }}-network.deb
 
       - name: Upload Artifact
         uses: actions/upload-artifact@v4
         with:
-          name: cortex-${{ inputs.new_version }}-linux-amd64-local-installer
+          name: cortex-${{ inputs.new_version }}-linux-${{ inputs.arch }}-local-installer
           path: ./engine/${{ steps.set-output-params.outputs.package_name }}-local.deb
 
       - name: upload to aws s3 if public provider is aws
         if: inputs.public_provider == 'aws-s3'
         run: |
-          aws s3 cp ./engine/cortex.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/linux-amd64-cortex-nightly.tar.gz
-          aws s3 cp ./engine/${{ steps.set-output-params.outputs.package_name }}-network.deb s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/cortex-linux-amd64-network-installer.deb
+          aws s3 cp ./engine/cortex.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/linux-${{ inputs.arch }}-cortex-nightly.tar.gz
+          aws s3 cp ./engine/${{ steps.set-output-params.outputs.package_name }}-network.deb s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/temp-latest/cortex-linux-${{ inputs.arch }}-network-installer.deb
 
-          aws s3 cp ./engine/cortex.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/v${{ inputs.new_version }}/linux-amd64/cortex-nightly.tar.gz
-          aws s3 cp ./engine/${{ steps.set-output-params.outputs.package_name }}-network.deb s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/v${{ inputs.new_version }}/linux-amd64/cortex-${{ inputs.new_version }}-linux-amd64-network-installer.deb
-          aws s3 cp ./engine/${{ steps.set-output-params.outputs.package_name }}-local.deb s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/v${{ inputs.new_version }}/linux-amd64/cortex-${{ inputs.new_version }}-linux-amd64-local-installer.deb
+          aws s3 cp ./engine/cortex.tar.gz s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/v${{ inputs.new_version }}/linux-${{ inputs.arch }}/cortex-nightly.tar.gz
+          aws s3 cp ./engine/${{ steps.set-output-params.outputs.package_name }}-network.deb s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/v${{ inputs.new_version }}/linux-${{ inputs.arch }}/cortex-${{ inputs.new_version }}-linux-${{ inputs.arch }}-network-installer.deb
+          aws s3 cp ./engine/${{ steps.set-output-params.outputs.package_name }}-local.deb s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/cortex/v${{ inputs.new_version }}/linux-${{ inputs.arch }}/cortex-${{ inputs.new_version }}-linux-${{ inputs.arch }}-local-installer.deb
         env:
           AWS_ACCESS_KEY_ID: ${{ secrets.DELTA_AWS_ACCESS_KEY_ID }}
           AWS_SECRET_ACCESS_KEY: ${{ secrets.DELTA_AWS_SECRET_ACCESS_KEY }}
@@ -223,7 +243,7 @@ jobs:
         with:
           upload_url: ${{ inputs.upload_url }}
           asset_path: ./engine/cortex.tar.gz
-          asset_name: cortex-${{ inputs.new_version }}-linux-amd64.tar.gz
+          asset_name: cortex-${{ inputs.new_version }}-linux${{ inputs.arch }}.tar.gz
           asset_content_type: application/zip
 
       - name: Upload release assert if public provider is github
@@ -234,7 +254,7 @@ jobs:
         with:
           upload_url: ${{ inputs.upload_url }}
           asset_path: ./engine/${{ steps.set-output-params.outputs.package_name }}-network.deb
-          asset_name: cortex-${{ inputs.new_version }}-linux-amd64-network-installer.deb
+          asset_name: cortex-${{ inputs.new_version }}-linux-${{ inputs.arch }}-network-installer.deb
           asset_content_type: application/octet-stream
 
       - name: Upload release assert if public provider is github
@@ -245,5 +265,5 @@ jobs:
         with:
           upload_url: ${{ inputs.upload_url }}
           asset_path: ./engine/${{ steps.set-output-params.outputs.package_name }}-local.deb
-          asset_name: cortex-${{ inputs.new_version }}-linux-amd64-local-installer.deb
+          asset_name: cortex-${{ inputs.new_version }}-linux-${{ inputs.arch }}-local-installer.deb
           asset_content_type: application/octet-stream
\ No newline at end of file
diff --git a/engine/Makefile b/engine/Makefile
index 8f27eebcc..a6a4b5a79 100644
--- a/engine/Makefile
+++ b/engine/Makefile
@@ -24,6 +24,7 @@ DESTINATION_BINARY_SERVER_NAME ?= cortex-server
 DATA_FOLDER_NAME ?= .cortex
 CONFIGURATION_FILE_NAME ?= .cortexrc
 UNINSTALLER_FILE_NAME ?= cortex-uninstall.sh
+ARCH ?= amd64
 
 # Default target, does nothing
 all:
@@ -120,7 +121,7 @@ else ifeq ($(shell uname -s),Linux)
 	@echo "Building installer for linux"; \
 	cd templates/linux; \
 	chmod +x create_deb.sh; \
-	./create_deb.sh $(PACKAGE_NAME) $(VERSION) $(SOURCE_BINARY_PATH) $(SOURCE_BINARY_SERVER_PATH) $(DESTINATION_BINARY_NAME) $(DESTINATION_BINARY_SERVER_NAME) $(DATA_FOLDER_NAME) $(CONFIGURATION_FILE_NAME); \
+	./create_deb.sh $(PACKAGE_NAME) $(VERSION) $(SOURCE_BINARY_PATH) $(SOURCE_BINARY_SERVER_PATH) $(DESTINATION_BINARY_NAME) $(DESTINATION_BINARY_SERVER_NAME) $(DATA_FOLDER_NAME) $(CONFIGURATION_FILE_NAME) $(ARCH); \
 	cp $(PACKAGE_NAME).deb ../../
 else
 	@echo "Building installer for Macos"; \
diff --git a/engine/e2e-test/test_api_engine.py b/engine/e2e-test/test_api_engine.py
index 57b47b879..e652e4495 100644
--- a/engine/e2e-test/test_api_engine.py
+++ b/engine/e2e-test/test_api_engine.py
@@ -28,14 +28,14 @@ def test_engines_get_llamacpp_should_be_successful(self):
         
     # engines install
     def test_engines_install_llamacpp_specific_version_and_variant(self):
-        data = {"version": "v0.1.35-27.10.24", "variant": "linux-amd64-avx-cuda-11-7"}
+        data = {"version": "v0.1.40-b4354", "variant": "linux-amd64-avx-cuda-11-7"}
         response = requests.post(
             "http://localhost:3928/v1/engines/llama-cpp/install", json=data
         )
         assert response.status_code == 200
 
     def test_engines_install_llamacpp_specific_version_and_null_variant(self):
-        data = {"version": "v0.1.35-27.10.24"}
+        data = {"version": "v0.1.40-b4354"}
         response = requests.post(
             "http://localhost:3928/v1/engines/llama-cpp/install", json=data
         )
diff --git a/engine/e2e-test/test_api_engine_update.py b/engine/e2e-test/test_api_engine_update.py
index 23939f038..be8685dba 100644
--- a/engine/e2e-test/test_api_engine_update.py
+++ b/engine/e2e-test/test_api_engine_update.py
@@ -25,7 +25,7 @@ def setup_and_teardown(self):
 
     @pytest.mark.asyncio
     async def test_engines_update_should_be_successfully(self):
-        requests.post("http://localhost:3928/v1/engines/llama-cpp?version=0.1.34")
+        requests.post("http://localhost:3928/v1/engines/llama-cpp?version=0.1.43")
         response = requests.post("http://localhost:3928/v1/engines/llama-cpp/update")
         assert response.status_code == 200
 
diff --git a/engine/e2e-test/test_api_model.py b/engine/e2e-test/test_api_model.py
index 8f2e4b07a..d75aa6831 100644
--- a/engine/e2e-test/test_api_model.py
+++ b/engine/e2e-test/test_api_model.py
@@ -85,6 +85,7 @@ async def test_model_pull_with_direct_url_should_have_desired_name(self):
             ],
         )
         
+    @pytest.mark.asyncio
     async def test_models_start_stop_should_be_successful(self):
         print("Install engine")
         response = requests.post("http://localhost:3928/v1/engines/llama-cpp/install")
diff --git a/engine/e2e-test/test_cli_engine_install.py b/engine/e2e-test/test_cli_engine_install.py
index a998f3183..dbbc16e8a 100644
--- a/engine/e2e-test/test_cli_engine_install.py
+++ b/engine/e2e-test/test_cli_engine_install.py
@@ -49,7 +49,7 @@ def test_engines_install_onnx_on_tensorrt_should_be_failed(self):
 
     @pytest.mark.skipif(platform.system() == "Windows", reason="Progress bar log issue on Windows")
     def test_engines_install_pre_release_llamacpp(self):
-        engine_version = "v0.1.29"
+        engine_version = "v0.1.43"
         exit_code, output, error = run(
             "Install Engine",
             ["engines", "install", "llama-cpp", "-v", engine_version],
@@ -69,7 +69,7 @@ def test_engines_install_pre_release_llamacpp(self):
         assert is_engine_version_exist, f"Engine version {engine_version} is not found"
         assert exit_code == 0, f"Install engine failed with error: {error}"
 
-    @pytest.mark.skipif(platform.system() == "Windows", reason="Progress bar log issue on Windows")
+    @pytest.mark.skipif(platform.system() == "Windows" or platform.system() == "Linux", reason="Progress bar log issue on Windows")
     def test_engines_should_fallback_to_download_llamacpp_engine_if_not_exists(self):
         exit_code, output, error = run(
             "Install Engine",
diff --git a/engine/e2e-test/test_cli_engine_install_nightly.py b/engine/e2e-test/test_cli_engine_install_nightly.py
index 8c66c284c..bbb56ac9b 100644
--- a/engine/e2e-test/test_cli_engine_install_nightly.py
+++ b/engine/e2e-test/test_cli_engine_install_nightly.py
@@ -47,6 +47,7 @@ def test_engines_install_onnx_on_tensorrt_should_be_failed(self):
         assert "is not supported on" in output, "Should display error message"
         assert exit_code == 0, f"Install engine failed with error: {error}"
 
+    @pytest.mark.skipif(platform.system() == "Linux", reason="Wait for linux arm ready")
     def test_engines_should_fallback_to_download_llamacpp_engine_if_not_exists(self):
         exit_code, output, error = run(
             "Install Engine",
diff --git a/engine/e2e-test/test_cli_engine_uninstall.py b/engine/e2e-test/test_cli_engine_uninstall.py
index fcc5f5c73..6b640d45d 100644
--- a/engine/e2e-test/test_cli_engine_uninstall.py
+++ b/engine/e2e-test/test_cli_engine_uninstall.py
@@ -24,7 +24,7 @@ def setup_and_teardown(self):
 
     @pytest.mark.asyncio
     async def test_engines_uninstall_llamacpp_should_be_successfully(self):
-        requests.post("http://127.0.0.1:3928/v1/engines/llama-cpp/install")
+        response = requests.post("http://localhost:3928/v1/engines/llama-cpp/install")
         await wait_for_websocket_download_success_event(timeout=None)
         exit_code, output, error = run(
             "Uninstall engine", ["engines", "uninstall", "llama-cpp"]
diff --git a/engine/services/download_service.cc b/engine/services/download_service.cc
index d855c8f61..a38dbe70e 100644
--- a/engine/services/download_service.cc
+++ b/engine/services/download_service.cc
@@ -140,10 +140,10 @@ cpp::result<uint64_t, std::string> DownloadService::GetFileSize(
   curl_easy_setopt(curl, CURLOPT_URL, url.c_str());
 
   auto headers = curl_utils::GetHeaders(url);
-  if (headers.has_value()) {
+  if (headers) {
     curl_slist* curl_headers = nullptr;
 
-    for (const auto& [key, value] : headers.value()) {
+    for (const auto& [key, value] : headers->m) {
       auto header = key + ": " + value;
       curl_headers = curl_slist_append(curl_headers, header.c_str());
     }
@@ -227,10 +227,10 @@ cpp::result<bool, std::string> DownloadService::Download(
 
   curl_easy_setopt(curl, CURLOPT_URL, download_item.downloadUrl.c_str());
   auto headers = curl_utils::GetHeaders(download_item.downloadUrl);
-  if (headers.has_value()) {
+  if (headers) {
     curl_slist* curl_headers = nullptr;
 
-    for (const auto& [key, value] : headers.value()) {
+    for (const auto& [key, value] : headers->m) {
       auto header = key + ": " + value;
       curl_headers = curl_slist_append(curl_headers, header.c_str());
     }
@@ -469,7 +469,7 @@ void DownloadService::SetUpCurlHandle(CURL* handle, const DownloadItem& item,
   auto headers = curl_utils::GetHeaders(item.downloadUrl);
   if (headers) {
     curl_slist* curl_headers = nullptr;
-    for (const auto& [key, value] : headers.value()) {
+    for (const auto& [key, value] : headers->m) {
       curl_headers =
           curl_slist_append(curl_headers, (key + ": " + value).c_str());
     }
diff --git a/engine/templates/linux/control b/engine/templates/linux/control
index e877fe5ab..7c129a690 100644
--- a/engine/templates/linux/control
+++ b/engine/templates/linux/control
@@ -2,7 +2,7 @@ Package: $PACKAGE_NAME
 Version: $VERSION
 Section: base
 Priority: optional
-Architecture: amd64
+Architecture: $ARCH
 Depends: openmpi-bin,libopenmpi-dev
 Maintainer: Homebrew Computer Pte Ltd <service@homebrew.ltd>
 Description: Cortex
diff --git a/engine/templates/linux/create_deb.sh b/engine/templates/linux/create_deb.sh
index 29492bdd8..f9247972f 100644
--- a/engine/templates/linux/create_deb.sh
+++ b/engine/templates/linux/create_deb.sh
@@ -6,6 +6,7 @@ DESTINATION_BINARY_NAME=$5
 DESTINATION_BINARY_SERVER_NAME=$6
 DATA_FOLDER_NAME=$7
 CONFIGURATION_FILE_NAME=$8
+ARCH=$9
 
 mkdir -p $PACKAGE_NAME/DEBIAN
 
@@ -31,8 +32,9 @@ chmod 755 $PACKAGE_NAME/DEBIAN/postinst
 chmod 755 $PACKAGE_NAME/DEBIAN/postrm
 chmod 755 $PACKAGE_NAME/DEBIAN/prerm
 
-export PACKAGE_NAME VERSION
+export PACKAGE_NAME VERSION ARCH
 
 envsubst < control > $PACKAGE_NAME/DEBIAN/control
+sed -i "s/ARCH/$ARCH/" $PACKAGE_NAME/DEBIAN/control
 
 dpkg-deb --build $PACKAGE_NAME $PACKAGE_NAME.deb
diff --git a/engine/templates/linux/create_deb_local.sh b/engine/templates/linux/create_deb_local.sh
index 6b54dc19d..0ab0c79c5 100644
--- a/engine/templates/linux/create_deb_local.sh
+++ b/engine/templates/linux/create_deb_local.sh
@@ -6,6 +6,7 @@ DESTINATION_BINARY_NAME=$5
 DESTINATION_BINARY_SERVER_NAME=$6
 DATA_FOLDER_NAME=$7
 CONFIGURATION_FILE_NAME=$8
+ARCH=$9
 
 mkdir -p $PACKAGE_NAME/DEBIAN
 
@@ -34,7 +35,7 @@ chmod 755 $PACKAGE_NAME/DEBIAN/postinst
 chmod 755 $PACKAGE_NAME/DEBIAN/postrm
 chmod 755 $PACKAGE_NAME/DEBIAN/prerm
 
-export PACKAGE_NAME VERSION
+export PACKAGE_NAME VERSION ARCH
 
 envsubst < control > $PACKAGE_NAME/DEBIAN/control
 
diff --git a/engine/templates/linux/install.sh b/engine/templates/linux/install.sh
index e11b879c6..ade0b134a 100644
--- a/engine/templates/linux/install.sh
+++ b/engine/templates/linux/install.sh
@@ -6,6 +6,17 @@ if [ "$(id -u)" != "0" ]; then
     exit 1
 fi
 
+# Determine architecture
+ARCH=$(uname -m)
+if [ "$ARCH" = "x86_64" ]; then
+    ARCH="amd64"
+elif [ "$ARCH" = "aarch64" ]; then
+    ARCH="arm64"
+else
+    echo "Unsupported architecture: $ARCH"
+    exit 1
+fi
+
 # Determine the home directory based on the user
 USER_TO_RUN_AS=${SUDO_USER:-$(whoami)}
 if [ "$USER_TO_RUN_AS" = "root" ]; then
@@ -142,19 +153,19 @@ install_cortex() {
 
   case $channel in
     stable)
-      url_binary="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-amd64.tar.gz"
-      url_deb_local="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-amd64-local-installer.deb"
-      url_deb_network="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-amd64-network-installer.deb"
+      url_binary="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-${ARCH}.tar.gz"
+      url_deb_local="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-${ARCH}-local-installer.deb"
+      url_deb_network="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-${ARCH}-network-installer.deb"
       ;;
     beta)
-      url_binary="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-amd64.tar.gz"
-      url_deb_local="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-amd64-local-installer.deb"
-      url_deb_network="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-amd64-network-installer.deb"
+      url_binary="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-${ARCH}.tar.gz"
+      url_deb_local="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-${ARCH}-local-installer.deb"
+      url_deb_network="https://github.com/janhq/cortex.cpp/releases/download/v${version}/cortex-${version}-linux-${ARCH}-network-installer.deb"
       ;;
     nightly)
-      url_binary="https://delta.jan.ai/cortex/v${version}/linux-amd64/cortex-nightly.tar.gz"
-      url_deb_local="https://delta.jan.ai/cortex/v${version}/linux-amd64/cortex-${version}-linux-amd64-local-installer.deb"
-      url_deb_network="https://delta.jan.ai/cortex/v${version}/linux-amd64/cortex-${version}-linux-amd64-network-installer.deb"
+      url_binary="https://delta.jan.ai/cortex/v${version}/linux-${ARCH}/cortex-nightly.tar.gz"
+      url_deb_local="https://delta.jan.ai/cortex/v${version}/linux-${ARCH}/cortex-${version}-linux-${ARCH}-local-installer.deb"
+      url_deb_network="https://delta.jan.ai/cortex/v${version}/linux-${ARCH}/cortex-${version}-linux-${ARCH}-network-installer.deb"
       ;;
   esac
 
diff --git a/engine/test/components/test_engine_matcher_utils.cc b/engine/test/components/test_engine_matcher_utils.cc
index 7da4e3cd1..1d1ed47a8 100644
--- a/engine/test/components/test_engine_matcher_utils.cc
+++ b/engine/test/components/test_engine_matcher_utils.cc
@@ -19,6 +19,7 @@ class EngineMatcherUtilsTestSuite : public ::testing::Test {
       "cortex.llamacpp-0.1.25-25.08.24-linux-amd64-noavx-cuda-12-0.tar.gz",
       "cortex.llamacpp-0.1.25-25.08.24-linux-amd64-noavx.tar.gz",
       "cortex.llamacpp-0.1.25-25.08.24-linux-amd64-vulkan.tar.gz",
+      "cortex.llamacpp-0.1.43-linux-arm64.tar.gz",
       "cortex.llamacpp-0.1.25-25.08.24-mac-amd64.tar.gz",
       "cortex.llamacpp-0.1.25-25.08.24-mac-arm64.tar.gz",
       "cortex.llamacpp-0.1.25-25.08.24-windows-amd64-avx-cuda-11-7.tar.gz",
@@ -134,6 +135,18 @@ TEST_F(EngineMatcherUtilsTestSuite, TestValidate) {
     EXPECT_EQ(variant,
               "cortex.llamacpp-0.1.25-25.08.24-windows-amd64-avx2.tar.gz");
   }
+
+  {
+    auto os{"linux"};
+    auto cpu_arch{"arm64"};
+    auto suitable_avx{""};
+    auto cuda_version{""};
+
+    auto variant = engine_matcher_utils::Validate(
+        cortex_llamacpp_variants, os, cpu_arch, suitable_avx, cuda_version);
+
+    EXPECT_EQ(variant, "cortex.llamacpp-0.1.43-linux-arm64.tar.gz");
+  }
 }
 
 TEST_F(EngineMatcherUtilsTestSuite, TestGetVersionAndArch) {
diff --git a/engine/utils/curl_utils.cc b/engine/utils/curl_utils.cc
index d5945e8c8..2481658ad 100644
--- a/engine/utils/curl_utils.cc
+++ b/engine/utils/curl_utils.cc
@@ -9,12 +9,24 @@
 
 namespace curl_utils {
 namespace {
-size_t WriteCallback(void* contents, size_t size, size_t nmemb,
-                     std::string* output) {
-  size_t totalSize = size * nmemb;
-  output->append((char*)contents, totalSize);
-  return totalSize;
-}
+class CurlResponse {
+ public:
+  static size_t WriteCallback(char* buffer, size_t size, size_t nitems,
+                              void* userdata) {
+    auto* response = static_cast<CurlResponse*>(userdata);
+    return response->Append(buffer, size * nitems);
+  }
+
+  size_t Append(const char* buffer, size_t size) {
+    data_.append(buffer, size);
+    return size;
+  }
+
+  const std::string& GetData() const { return data_; }
+
+ private:
+  std::string data_;
+};
 
 void SetUpProxy(CURL* handle, const std::string& url) {
   auto config = file_manager_utils::GetCortexConfig();
@@ -59,19 +71,18 @@ void SetUpProxy(CURL* handle, const std::string& url) {
 }
 }  // namespace
 
-std::optional<std::unordered_map<std::string, std::string>> GetHeaders(
-    const std::string& url) {
+std::shared_ptr<Header> GetHeaders(const std::string& url) {
   auto url_obj = url_parser::FromUrlString(url);
   if (url_obj.has_error()) {
-    return std::nullopt;
+    return nullptr;
   }
 
   if (url_obj->host == kHuggingFaceHost) {
-    std::unordered_map<std::string, std::string> headers{};
-    headers["Content-Type"] = "application/json";
+    auto headers = std::make_shared<Header>();
+    headers->m["Content-Type"] = "application/json";
     auto const& token = file_manager_utils::GetCortexConfig().huggingFaceToken;
     if (!token.empty()) {
-      headers["Authorization"] = "Bearer " + token;
+      headers->m["Authorization"] = "Bearer " + token;
 
       // for debug purpose
       auto min_token_size = 6;
@@ -87,15 +98,15 @@ std::optional<std::unordered_map<std::string, std::string>> GetHeaders(
   }
 
   if (url_obj->host == kGitHubHost) {
-    std::unordered_map<std::string, std::string> headers{};
-    headers["Accept"] = "application/vnd.github.v3+json";
+    auto headers = std::make_shared<Header>();
+    headers->m["Accept"] = "application/vnd.github.v3+json";
     // github API requires user-agent https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api?apiVersion=2022-11-28#user-agent
     auto user_agent = file_manager_utils::GetCortexConfig().gitHubUserAgent;
     auto gh_token = file_manager_utils::GetCortexConfig().gitHubToken;
-    headers["User-Agent"] =
+    headers->m["User-Agent"] =
         user_agent.empty() ? kDefaultGHUserAgent : user_agent;
     if (!gh_token.empty()) {
-      headers["Authorization"] = "Bearer " + gh_token;
+      headers->m["Authorization"] = "Bearer " + gh_token;
 
       // for debug purpose
       auto min_token_size = 6;
@@ -109,7 +120,7 @@ std::optional<std::unordered_map<std::string, std::string>> GetHeaders(
     return headers;
   }
 
-  return std::nullopt;
+  return nullptr;
 }
 
 cpp::result<std::string, std::string> SimpleGet(const std::string& url,
@@ -122,8 +133,8 @@ cpp::result<std::string, std::string> SimpleGet(const std::string& url,
 
   auto headers = GetHeaders(url);
   curl_slist* curl_headers = nullptr;
-  if (headers.has_value()) {
-    for (const auto& [key, value] : headers.value()) {
+  if (headers) {
+    for (const auto& [key, value] : headers->m) {
       auto header = key + ": " + value;
       curl_headers = curl_slist_append(curl_headers, header.c_str());
     }
@@ -131,12 +142,14 @@ cpp::result<std::string, std::string> SimpleGet(const std::string& url,
     curl_easy_setopt(curl, CURLOPT_HTTPHEADER, curl_headers);
   }
 
-  std::string readBuffer;
+  auto* response = new CurlResponse();
+  std::shared_ptr<CurlResponse> s(response,
+                                  std::default_delete<CurlResponse>());
 
   SetUpProxy(curl, url);
   curl_easy_setopt(curl, CURLOPT_URL, url.c_str());
-  curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, WriteCallback);
-  curl_easy_setopt(curl, CURLOPT_WRITEDATA, &readBuffer);
+  curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, CurlResponse::WriteCallback);
+  curl_easy_setopt(curl, CURLOPT_WRITEDATA, response);
   if (timeout > 0) {
     curl_easy_setopt(curl, CURLOPT_TIMEOUT, timeout);
   }
@@ -155,10 +168,10 @@ cpp::result<std::string, std::string> SimpleGet(const std::string& url,
   if (http_code >= 400) {
     CTL_ERR("HTTP request failed with status code: " +
             std::to_string(http_code));
-    return cpp::fail(readBuffer);
+    return cpp::fail(response->GetData());
   }
 
-  return readBuffer;
+  return response->GetData();
 }
 
 cpp::result<std::string, std::string> SimpleRequest(
@@ -176,13 +189,15 @@ cpp::result<std::string, std::string> SimpleRequest(
       curl_slist_append(curl_headers, "Content-Type: application/json");
   curl_headers = curl_slist_append(curl_headers, "Expect:");
 
-  if (headers.has_value()) {
-    for (const auto& [key, value] : headers.value()) {
+  if (headers) {
+    for (const auto& [key, value] : headers->m) {
       auto header = key + ": " + value;
       curl_headers = curl_slist_append(curl_headers, header.c_str());
     }
   }
-  std::string readBuffer;
+  auto* response = new CurlResponse();
+  std::shared_ptr<CurlResponse> s(response,
+                                  std::default_delete<CurlResponse>());
 
   SetUpProxy(curl, url);
   curl_easy_setopt(curl, CURLOPT_HTTPHEADER, curl_headers);
@@ -196,8 +211,8 @@ cpp::result<std::string, std::string> SimpleRequest(
     curl_easy_setopt(curl, CURLOPT_CUSTOMREQUEST, "DELETE");
   }
   curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1L);
-  curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, WriteCallback);
-  curl_easy_setopt(curl, CURLOPT_WRITEDATA, &readBuffer);
+  curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, CurlResponse::WriteCallback);
+  curl_easy_setopt(curl, CURLOPT_WRITEDATA, response);
 
   curl_easy_setopt(curl, CURLOPT_POSTFIELDSIZE, body.length());
   curl_easy_setopt(curl, CURLOPT_POSTFIELDS, body.c_str());
@@ -221,10 +236,10 @@ cpp::result<std::string, std::string> SimpleRequest(
   if (http_code >= 400) {
     CTL_ERR("HTTP request failed with status code: " +
             std::to_string(http_code));
-    return cpp::fail(readBuffer);
+    return cpp::fail(response->GetData());
   }
 
-  return readBuffer;
+  return response->GetData();
 }
 
 cpp::result<YAML::Node, std::string> ReadRemoteYaml(const std::string& url) {
diff --git a/engine/utils/curl_utils.h b/engine/utils/curl_utils.h
index f33b7e8e5..9035b6b3c 100644
--- a/engine/utils/curl_utils.h
+++ b/engine/utils/curl_utils.h
@@ -15,8 +15,11 @@ enum class RequestType { GET, PATCH, POST, DEL };
 
 namespace curl_utils {
 
-std::optional<std::unordered_map<std::string, std::string>> GetHeaders(
-    const std::string& url);
+struct Header {
+  std::unordered_map<std::string, std::string> m;
+};
+
+std::shared_ptr<Header> GetHeaders(const std::string& url);
 
 cpp::result<std::string, std::string> SimpleGet(const std::string& url,
                                                 const int timeout = -1);
diff --git a/engine/utils/engine_matcher_utils.h b/engine/utils/engine_matcher_utils.h
index a6135e532..28c0f0c2a 100644
--- a/engine/utils/engine_matcher_utils.h
+++ b/engine/utils/engine_matcher_utils.h
@@ -156,6 +156,11 @@ inline std::string Validate(const std::vector<std::string>& variants,
   if (os == "mac" && !os_and_arch_compatible_list.empty())
     return os_and_arch_compatible_list[0];
 
+  if (os == "linux" && cpu_arch == "arm64" &&
+      !os_and_arch_compatible_list.empty()) {
+    return os_and_arch_compatible_list[0];
+  }
+
   std::vector<std::string> avx_compatible_list;
 
   std::copy_if(os_and_arch_compatible_list.begin(),